Every video needed the same eight jobs
We started a short-video channel that teaches how AI and tech actually work: why a detector flags plain English, how an image model turns static into a picture, why the same question gets two answers. Each video has to explain one mechanism in under a minute, with a sourced number and a hook in the first frame.
Done by hand, every video meant the same eight jobs: script, voice, word timings, captions, scenes, music, layout checks and a cover. That is the work that stops a series after episode three, and it is the same work a brand faces when it wants explainers every week instead of one launch film a year.
One command from script to finished video
We wrote a pipeline that takes a short spec and a script and returns a finished vertical video. One command makes the AI voiceover with word timings, syncs captions to every word, renders the scenes as HTML motion graphics or realistic Three.js 3D, adds music and sound effects, checks the layout against each platform's safe zones and renders the video and its cover.
When a person is on camera, the pipeline cleans up their own voice and aligns every word with Whisper; it never replaces a real voice with an AI one. Every look we have built is kept in a style catalogue, so a series keeps one visual identity from the first episode to the fortieth.
The rules come from watching what viewers did, not from a brainstorm: an event in the first frame in six words or fewer, one idea per scene, plain numbered steps, slow layout moves, and sources in the comments rather than over the video.
A series instead of a one-off
Video count from our own production log. Views from our Instagram dashboard for the 30 days to 30 October 2026, read on that day. These are small numbers from a new account: the result that matters here is the production rate.
What we would do next
Client batches: the same pipeline makes product explainers, onboarding videos and launch films from a client's script and brand kit, one command per video. It is sold the same way as any AI content work, starting with a pilot batch.
This is our own tool, so nothing here is a client commitment. It is where we would take the same build for a client who wants one like it.
Team on this project
The same people who scoped it built it. No handover to a delivery pod.
Related work
Questions about this project
On explainers without a presenter, yes: an ElevenLabs voice with word timings. When a person is on camera, it is their own voice, cleaned up and never replaced. Videos that use an AI voice or AI images carry the platform's AI label.
Music from a CC0 library, so there is nothing to license, and sound effects made in code from the video's own timing. A client can supply licensed brand music instead.
Yes. Each series gets one style key: type, colours, layout, where the presenter sits and how scenes move. The catalogue already holds seven looks to start from.
Not by default. Scenes are built in code from free and licensed assets, which keeps the cost per video low and predictable. Paid generated footage is used only when a client approves the cost for a specific shot.